Articles

Stroke-unit care for acute stroke patients: an observational

follow-up study
Livia Candelise, Monica Gattinoni, Anna Bersano, Giuseppe Micieli, Roberto Sterzi, Alberto Morabito, on the behalf of the PROSIT Study Group

Summary
Background Large numbers of stroke patients arrive at hospital at a very early stage, and eective treatments for the Lancet 2007; 369: 299305
acute phase of the disease are available. However, evidence that patients with acute stroke benet from stroke-unit See Comment page 254
care is scarce. Our aim was to determine whether admission to a stroke unit, rather than a conventional ward, aected Dipartimento di Scienze
the outcome of patients with acute stroke. Neurologiche, Ospedale
Maggiore Policlinico IRCCS,
Universit degli Studi di
Methods We did an observational follow-up study of 11 572 acute stroke patients hospitalised within 48 h of the onset Milano, Milan, Italy
of symptoms either in a stroke unit (n=4936) or in a conventional ward (6636). Patients were identied retrospectively (Prof L Candelise MD,
from discharge records from 260 Italian hospitals. The primary outcome was mortality or disability (Rankin score M Gattinoni PhD,
A Bersano MD); Stroke Unit,
greater than two), assessed prospectively by independent, masked assessors 2 years after admission. Analyses were Istituto Neurologico Mondino
adjusted for patient characteristics and clustered at the hospital level. IRCCS, Pavia, Italy
(G Micieli MD); SC Neurologia,
Findings Overall, 1576 patients died in hospital; 2169 died during the follow-up period. 347 patients were lost to Ospedale Niguarda CaGranda,
Milano, Milan, Italy
follow-up. Compared with conventional-ward care, stroke-unit care was associated with a reduced probability of death (R Sterzi MD); and Cattedra di
or being disabled at the end of follow-up (odds ratio 081, 95% CI 072091; p=00001). The potential benet was Statistica Medica, Facolt di
signicant across all age ranges and clinical characteristics, except for unconsciousness. No specic elements of Medicina e Chirurgia,
Universit degli Studi di
setting, organisation, or process of care were associated with outcome.
Milano, Milan, Italy
(Prof A Morabito MD)
Interpretation Admission to a stroke-unit ward with dedicated beds and sta within 48 h of onset should be Correspondence to:
recommended for all patients with acute stroke. Prof Livia Candelise,
Dipartimento di Scienze
Introduction 2000 and 2004, to compare the long-term outcome of Neurologiche, Universit degli
Studi di Milano, Ospedale
Several randomised controlled trials have been done in patients cared for in stroke units with those cared for in Maggiore Policlinico, Via F
dierent countries and environments to assess the eective- conventional wards. Seven Italian regions (Lombardia, Sforza, 35, 20122 Milano, Italy
ness of stroke-unit care for hospitalised stroke patients.1 Liguria, Veneto, Friuli-Venezia Giulia, Emilia-Romagna, livia.candelise@unimi.it

The results of these studies were combined in a Cochrane
systematic review, which denitively showed the benet of
stroke-unit care.1 However, there is still some delay in
implementation of stroke-unit wards in European countries.2
The large-scale applicability of some of the complex
interventions that were used experimentally on small,
selected groups of patients included in the randomised
trials is debatable. Guidelines dier about important aspects
of stroke careeg, acuteness for admission, patient
selection, and care organisation.3 In North America there is
even disagreement on the need to have a discrete, dedicated
ward together with a dedicated stroke team.4,5
The Italian National Ministry of Health has promoted a
stroke-unit research project (Research Project on Stroke
services in Italy, PROSIT) since 2000, which involves
many institutions around the country. This project enabled
us to do a large-scale controlled prospective trial to assess
the eectiveness of stroke-unit care. Our aim was to
determine whether all acute stroke patients admitted into
stroke units have a better long-term outcome than those
treated in conventional wards.
Methods
Participants and procedures
This study is an observational controlled follow-up study
on acute stroke patients, done in hospitals in Italy between
Toscana, and Lazio), with a population of about 29 million
and more than 50 000 acute stroke discharges every year,
took part in the study.
The study was done in accordance with guidelines for
good clinical practice and the declaration of Helsinki.
Central and local ethical committee approval was sought.
A patient information standard procedure supported
the patient consent process to use clinical data for
research purposes. Oral consent was obtained by phone
at the time of the follow-up telephone call.
We dened the stroke unit as a hospital ward with
dedicated beds (at least 80% stroke admission) and with
a dedicated stroke sta (at least one physician and one
nurse) who work exclusively in the care of stroke
patients. The conventional ward was dened as a
hospital service with neither dedicated beds nor sta, in
which stroke patients are cared for together with other
patients (eg, neurological, internal medicine, cardio-
logical). To identify stroke units and conventional
wards, we audited the local hospital services of all
regional hospital sites that had discharged at least
50 acute stroke patients in the previous year. We also
gathered data on the service characteristics (setting,
organisation, and process of care). The audit process
terminated in February, 2001, and 31 stroke units and
393 conventional wards were identied. The details of

www.thelancet.com Vol 369 January 27, 2007 299

 Articles
the data gathered during this phase of the study have
been described previously.6,7
For ward characteristics, the major elements were
dichotomised as follows: hospital setting (academic
hospital, presence in the same hospital of intensive
care, neurosurgery, vascular surgery, and rehabilitation
departments), unit setting (all neurological beds, yearly
unit admission more than 100 patients per year,
continuous electrocardiography monitoring equipment
more than two per ten beds), stang per unit ward
(number of full-time physicians, nurses, and therapists:
more than 2, 5, and 05 per ten patients, respectively),
diagnostic investigation (availability over 24 h, 7 days a
week, of cranial CT scan, brain MRI scan,
echocardiography, duplex ultrasound, and cerebral
catheter angiography), and process of care (declared
use in the unit of written nurse care protocols, practice
of multidisciplinary team meetings, and of early
mobilisation).
Acute stroke patients were identied by the Discharge
Related Group (DRG) 14 code. DRG 14 corresponds to
ICD-9-CM 431, 432, 433, 434, 436, and 437 codes that have
been shown to be highly specic for acute stroke.8
Ischaemic and haemorrhagic strokes were accepted. The
diagnosis and the duration of symptoms (less than 48 h)
were conrmed by assessment of clinical records. We
included all cases with acute stroke who were admitted
during the last 9 months of 2001 in the 31 stroke units
previously identied. Patients for whom clinical data were
not available or those readmitted to the same service in
1 year were excluded. Controls were a sample selected with
a computer-generated random numbers list stratied by
regions of about 20% of acute stroke cases admitted in
each of the 393 conventional wards during the same
period. To avoid selection and performance bias, we
31 stroke units in 30 hospitals* 393 conventional wards in
244 hospitals*
5604 consecutive admissions 7092 randomly sampled from
in 2001 eligible 2001 admissions eligible
332 unavailable data (6%) 158 unavailable data (2%)
152 readmissions (3%) 85 readmissions (1%)
184 refused consent (3%) 213 refused consent (3%)
4936 stroke-unit patients with 6636 control patients with
data at baseline data at baseline
172 lost to follow-up 175 lost to follow-up
4764 stroke-unit patients 6461 control patients
completed follow-up completed follow-up
Figure 1: Trial prole
*14 hospitals had both stroke unit and conventional ward.
300
included the patients admitted after nishing the auditing
process described above.
16 trained independent clinicians reviewed patients
clinical records blinded to the patients hospital status.
They conrmed diagnoses of acute stroke and recorded
demographic, clinical, and in-hospital outcome data. The
following baseline characteristics were recorded: age, male
sex, admission time, intracranial haemorrhage at rst CT
scan, atrial brillation at rst electrocardiograph, systolic
and diastolic arterial blood pressure, unconsciousness
(score less than 4 of eye opening on the Glasgow coma
scale) and drift in at least one limb at motor assessment.
Follow-up assessments were done between 2002 and
2004 centrally for all regions. The assessors were an
independent group of 21 medical students and four nurses
trained to use a structured follow-up form, masked to the
patients hospital status. They registered the patients vital
status, place of residence, and functional ability with a
modied Rankin scale. The modied Rankin scale graded
the patients from grade 0 (no symptoms) to grade 5
(bedridden, incontinent, and requiring constant nursing
care and attention). The inter-observer agreement and its
use by telephone are reliable.9,10 If the patient was unwilling
to answer the questions, information was obtained from a
carer or proxy. About a third of patients38% of those
receiving stroke-unit care and 32% of controlscompleted
the telephone interviews by themselves.
The primary outcome measure was death or dependency
condition (ie, Rankin score higher than two) recorded at
the time of long-term follow-up. Secondary outcomes were
in-hospital case fatality, long-term mortality, and long-term
living condition (ie, whether the patient was living at home
or not) at follow-up.
Statistical analysis
We aimed to recruit a sample of about 10 000 cases from
about 200 hospitals, each admitting at least ten patients,
which would have 80% power to detect an absolute risk
dierence of 6% (two sided p=005) if 62% of those
treated in a conventional ward were dead or had a poor
outcome, to detect an absolute risk dierence of 3% if
18% treated in conventional wards were dead at discharge,
and to detect an absolute risk dierence of 3% if 29%
treated in a conventional ward were not living at home at
the long-term follow-up assessment. The sample size for
hospital clusters was estimated assuming a coecient of
variation of 022.11
Hospitals intra-cluster correlation coecients were
calculated for all baseline variables.12 The outcome
analyses take into account all included cases. Cases lost to
follow-up were counted in the denominators. To
determine whether their inclusion aected the results,
we did best-case and worst-case scenario analyses. The
best-case scenario was analysed on the basis of the
assumption that the patients who were lost to follow-up
were alive and independent in the stroke-unit group and
dead or disabled in the conventional ward group. We also
www.thelancet.com Vol 369 January 27, 2007
 Articles

did a worst-case scenario analysis on the basis of the to hospitals without such units showed much the same
assumption that the patients lost to follow-up in the stroke association (081, 071091; p=00001).
unit were dead or disabled and were alive and independent The possible benet of stroke-unit care was also evident
in the conventional ward group. for in-hospital, long-term mortality, and for the likelihood
Multivariate analyses were done to assess the association of not being at home 2 years after hospital discharge
between type of unit care and patient outcomes, adjusting (table 3). Survival of patients treated in stroke units was
for patient characteristics (age, sex, time from stroke onset,
intracranial haemorrhage, atrial brillation, level of Patients in stroke Patients in Intra-class
consciousness). All multivariate logistic regression models unit (n=4936) control wards correlation
(n=6636) coecient
for the main and subgroup analyses were done with
generalised estimating equations to account for clustering Age (years) 72 (129) 76 (122) 0038
of predictor variables and outcomes at dierent hospital Men 2590 (52%) 3195 (47%) 0001
levels. To determine goodness-of-t, the nal model was Admission within 6 h 1926 (39%) 2526 (36%) 0168
assessed by Hosmer-Lemeshow test with ten risk intervals. Intracranial haemorrhage 412 (7%) 859 (13%) 0214
For time-to-event outcomes, we used Cox regression Atrial brillation 794 (16%) 1280 (19%) 0034
survival analyses; hazard ratios were clustered at hospital Systolic blood pressure (mm Hg) 159 (289) 164 (374) 0022
level. Statistical analyses were done with Stata version 8.0. Diastolic blood pressure (mm Hg) 87 (144) 90 (144) 0043
Unconsciousness 675 (13%) 1303 (20%) 0034
Role of the funding source Unconsciousness or motor impairment 3297 (70%) 4576 (69%) 0066
The Italian National Ministry of Health and Lombardia Aphasia 1307 (25%) 1819 (26%) 0042
Regional Health Department funded the study. The Length of stay in hospital (days) 12 (113) 12 (122) 0070
sponsors had no role in study design, data collection, data
*Adjusted for hospital clusters. Data are mean (SD) or number (%).
analysis, data interpretation or writing the report. The
corresponding author had full access to all the data, and Table 1: Distribution of baseline characteristics
had nal responsibility to submit for publication.
Stroke unit (n=4936) Control (n=6636)
Results
Follow-up (months) 197 (69) 204 (72)
Figure 1 shows the trial prole. At baseline, data were
available for 11 572 acute stroke patients from 424 units and Lost to follow-up 172 (3%) 175 (3%)

260 hospitals were studied. 16 hospitals had only stroke In-hospital case fatality 542 (11%) 1034 (16%)

units and 230 only conventional wards; 14 hospitals had Death after discharge 821 (17%) 1348 (20%)
both types of ward. The number of cases for the three types Alive at follow-up 3401 (69%) 4079 (61%)
of hospitals were 2502, 6260, and 2810, respectively. Table 1 Rankin score=0* 735 (22%) 804 (20%)
shows the baseline characteristics for these patients. 90% Rankin score=1* 871 (26%) 941 (23%)
of baseline data were complete. Rankin score=2* 547 (16%) 604 (15%)
Table 2 presents data on the follow-up status of patients. Rankin score=3* 590 (17%) 740 (18%)
95% of follow-up data were complete. 347 (3%) patients Rankin score=4* 471 (14%) 713 (17%)
were not found at the telephone number provided during Rankin score=5* 187 (5%) 277 (7%)
hospitalisation, and it was not possible to trace them by Stroke recurrence 195 (4%) 265 (4%)
asking the local municipal administrative oce of declared Rehabilitation programme 1089 (22%) 1381 (21%)
residence. These individuals were thus lost to follow-up. New hospital admissions 835 (17%) 992 (15%)
These cases did not dier signicantly from the included
Data are mean (SD) or number (%).*Data are number (percentage of those alive at
cases in terms of the characteristics of patients. The mean follow-up).
follow-up time was much the same for stroke units as it
was control groups (197 [SD 69] months vs 204 [72] Table 2: Status of patients 2 years after admission to hospital

months). Overall, 1576 patients died in the hospital; 2169

(19%) died during the follow-up period (table 2). Stroke unit Control Odds ratio p value
Patients who received stroke-unit care were less likely to (n=4936) (n=6636) (95% CI)*
be dead or disabled than the controls by the end of follow- In-hospital case fatality 542 (11%) 1034 (15%) 078 (064095) 0016
up after adjusting for inter-hospital variability (OR 081,
Long-term mortality 1363 (28%) 2382 (36%) 079 (068091) 0001
95% CI 072091; p=00001; table 3). Inclusion of those
Death or disability 2611 (53%) 4112 (62%) 081 (072091) 00001
lost to follow-up did not aect our results; OR in the best-
case scenario was 070 (95% CI 061079), whereas in the Not living at home 1743 (35%) 2829 (3%) 085 (074097) 0019
worst-case scenario it was 091 (082099). Comparison *Adjusted by age, sex, time from stroke onset, intracranial haemorrhages, atrial brillation, and unconsciousness, and
of the outcomes of patients admitted to hospitals with clustered at the hospital level.
stroke units, irrespective of whether they were admitted to
Table 3: Associations between type of service
stroke unit or normal ward, with those of patients admitted

www.thelancet.com Vol 369 January 27, 2007 301

 Articles

signicantly higher than of controls (hazard ratio 086,

100 077097; p=00001; gure 2). The dierence in survival
between the two groups was most pronounced during
Stroke unit the rst month after admission (gure 2).
75 Stroke-unit care was associated with a reduction in the
likelihood of death or disability for all subgroups of
Survival (%)

Conventional patients. There were no signicant interactions between

50 ward patient characteristics and the eect of stroke-unit care,
with the exception of patients with intracranial
haemorrhage, who did better if cared for in a stroke-unit
25 ward (gure 3).
None of the 19 elements of setting, organisation, and
processes of care in stroke units changed the probability
0 of the patient being dead or disabled by the end of follow-
0 12 24 up, except for MRI scan availability 24 h a day, 7 days a
Time (months)
Number at risk
week. Patients admitted to a conventional ward had a
Stroke unit 6636 4709 4398 better outcome if the ward beds were neurological and
Conventional ward 4936 3859 3649 belonged to an academic hospital (gure 4).

Figure 2: Survival curves for patients admitted to stroke unit or conventional ward
Discussion
This observational follow-up study, done in the clinical
environment in Italy on a large sample of patients with
Stroke unit Conventional ward Odds ratio
acute stroke, shows that there is an association between
stroke-unit care provided in the acute phase of the disease
n N n N 95% CI
and improvement of stroke outcome. In this setting,
Overall 2611 4936 4112 6636 stroke-unit care could reduce in-hospital case fatality and
also increase the proportion of patients living independently
Age
Under 75 years 855 2513 1097 2758 about 2 years after discharge. The possible benet is
Over 75 years 1756 2423 3033 3878 maintained over time and is evident both for patients aged
OR (95%) interaction
105 (089125) less than 75 years and for those over 75 years.
Our results are in agreement with those of a meta-
Sex
Women 1390 2346 2358 3441
analysis of 23 randomised controlled trials on about
Men 1221 2590 1754 3195 5000 participants.1 We also provide evidence that acute
OR (95%) interaction stroke-unit care (ie, admitting patients within 48 h of
098 (082116)
stroke onset) could improve the patients outcome. Other
Time of admission evidence for this eect is scarce, despite the fact that most
Within 6 h 1113 1926 1631 2526
After 6 h 1498 3009 2481 4110
new stroke units had been organised for the care of very
OR (95%) interaction acute patients. In fact, 13 systematic reviews of this area of
091 (076109) research included participants within 7 days of stroke
Intracranial haemorrhage onset, nine included patients after 7 days, and only one
Yes 157 412 575 859 trial enrolled patients within 24 h of stroke. This small
No 2454 4524 3537 5777
OR (95%) interaction
trial was unable alone to show any benet for stroke-unit
156 (116210) care (OR 094, 95% CI 067133).13
The design of our study did not allow a random,
Atrial brillation
Yes 562 794 976 1280 concealed assignment of patients to the two treatment
No 2049 4142 3136 5356 modalities, and the apparent benets for stroke-unit care
OR (95%) interaction
098 (077124) could be due to selection bias and incomplete adjustment
for patient characteristics. However, our ndings are lent
Consciousness
Unconscious 601 675 1191 1303
support by several factors. First, the directors and sta of
Conscious 2010 4261 2921 5333 the stroke units and conventional wards studied here were
OR (95%) interaction unaware of the ongoing trial when they admitted and
089 (064124)
05 1 15
treated the enrolled patients, which should reduce
Favours Favours selection and performance bias. Second, all major known
stroke unit conventional ward outcome predictors that had an unbalanced distribution
between the two treatment groups were included in the
Figure 3: Eect of stroke unit care on death or disability by patient subgroups logistic regression analysis. The analysis model had a high
Data adjusted for patient characteristics and clustered at the hospital level. goodness-of-t, and the inclusion of other, less relevant

302 www.thelancet.com Vol 369 January 27, 2007

 Articles

predictors would not have modied the nal analysis

Stroke unit Control
result substantially (Hosmer-Lemeshow 2 490; df 10;
p>07678). Third, the analysis that compared outcomes of Odds ratio (95% CI) Odds ratio (95% CI)
patients admitted to hospitals with stroke units to those Hospital setting
Intensive unit 099 (065152) 118 (094141)
admitted to hospitals without stroke units conrmed the Neurosurgery department 127 (084189) 099 (083119)
benet of treatment in a stroke unit. Such intra-hospital Rehabilitation department 118 (084189) 097 (085110)
Vascular surgery department 105 (079137) 118 (102137)
analysis should restrict the error due to in-hospital Academic hospital 123 (095159) 072 (058090)
selection bias. Some uncertainty of the validity of our
results still remains due to the non-randomised nature of Unit setting
Neurological beds 088 (055139) 064 (055075)
this study. However, to do a randomised trial would be More than 100 patients/year 114 (075172) 091 (079104)
impossible for ethical reasons. Moreover, assessment of More than two monitors/bed 101 (078133) 086 (054137)
outcome in our study was prospective and masked, and Stang
less than 3% of patients were lost to follow-up. Additionally, More than 2 physicians/10 pts 076 (047122) 093 (081108)
the analysis was clustered at the hospital level to account More than 5 nurses/10 pts 101 (073141) 102 (08915)
More than 05 therapists/10 pts 101 (076133) 106 (091125)
for service variability. Furthermore, the inclusion of a large
sample of stroke patients means that our ndings should Process of care
Nurses protocols 082 (051132) 093 (083106)
have a good level of generalisability. Early mobilisation 089 (051132) 093 (082105)
Other non-randomised prospective studies that aimed Multidisciplinary meetings 097 (075125) 099 (088112)
to assess the eect of stroke-unit care have produced
Diagnostic exams available 24/7
conicting results. The Swedish Riks-Stroke studies14,15 CT scan 154 (066357) 081 (062106)
suggest that stroke-unit care is favourable only for less MRI scan 069 (051089) 094 (082109)
Ultrasonography 135 (096189) 093 (081106)
severe patients, whereas the UK national stroke audit16 Echocardiography 081 (060111) 098 (086111)
showed that stroke-unit care substantially reduced in- Digital angiography 076 (049116) 096 (083111)
hospital case fatality, and the US Brain Attack Coalition
05 1 15 05 1 15
study17 did not show any short-term benet of treatment in
a stroke unit. Dierences in study design could explain, in Favour service characteristics
part, the diering results. Both the UK and US studies
considered short-term outcomes; the eects of stroke-unit Figure 4: Eect of service characteristics on death or disability
care seem to be more apparent after long-term follow-up, Stroke unit and control services were analysed separately. Data adjusted for patient and unit characteristics and
clustered at the hospital level.
as we have shown here.1,18 However, higher standards of
care for stroke in the conventional-ward groups in previous
studies could have reduced their ability to identify exactly what other components of those services could be
dierences in outcomes compared with stroke units. In crucial for achieving better outcomes. That we showed
fact, more than 70% of hospitals included in these three that none of the major stroke-unit elements of care was
studies had an organised stroke unit.1517 The presence of a independently associated with a better outcome is
stroke unit probably provides a general stimulus for stroke unsurprising. However, caution is needed when
management, inuencing the quality of stroke care even interpreting subgroup analyses, especially since only a
in wards with no formal stroke-unit organisation. By small number of outcome events were observed, which
contrast, only 12% of hospitals in the seven Italian regions restricted the statistical power. Nonetheless, our ndings
studied here provided stroke-unit care. are in line with those of the Brain Attack Coalition study,17
To be able to accurately predict which patients will gain which used a similar multifactorial analysis, and in part
most from stroke-unit care could be important, not just with those of the SCOPES study,20 which recorded only a
because of restricted resources, but also to aid individual slight association between the probability of being
patients to get the most appropriate access. Some discharged and complete adherence to 15 process-of-care
policymakers suggest the use of pragmatically restricted rules. Other studies considered either only the association
admission policies. However, this study, in accordance between elements of care with a stroke unit15,2123 or the
with other evidence, lends support to the view that there eect on outcome of separate elements such as presence
is no rm evidence for restricting access according to of a continuous monitoring system24,25 or the practice of
patient characteristicsie, age, sex, blood pressure, etc.1,19 early rehabilitation.26 The eectiveness of a single element
Previous reports,1,14 together with our results, indicate of care could be determined only by doing a large
restricted use of stroke-unit care for patients with mild randomised controlled trial in which the other elements
stroke. However, a larger sample of unconscious patients do not change during the experiment, which would be
would be necessary to conrm this nding. very dicult. Nevertheless, the overall eectiveness of
We aimed to identify the key components of an eective stroke-unit care could also be attributed to non-measurable
acute stroke service. In view of the emergent evidence factorseg, the enthusiasm, motivation, and
that a stroke unit with a minimum of dedicated beds and specialisation of stawhich cannot be attributed to a
a dedicated sta is eective, a key goal was to determine single individual element separately.

www.thelancet.com Vol 369 January 27, 2007 303

 Articles
In conclusion, this study, which used a broad but
reliable denition of stroke-unit models on the basis of
the use of beds and sta dedicated exclusively to stroke
patients, indicates that, provided admission occurs within
48 h of onset, patients with acute stroke should be treated
in stroke-unit wards with dedicated beds and sta,
irrespective of age. At present, there is no indication that
any further element of care is of benet. This simple
organisation of services is easy to implement, with great
benet in terms of reduced mortality and disability.
PROSIT study group
Coordinating CenterL Candelise (Ospedale Maggiore Policlinico
IRCCS, Milano), G Micieli (IRCCS Fondazione Istituto Neurologico
C Mondino, Pavia), R Sterzi (SC Neurologia, Ospedale Niguarda
CaGranda, Milano, M Gattinoni Ospedale Maggiore Policlinico IRCCS,
Milano).
Steering CommitteeL Candelise (Ospedale Maggiore Policlinico
IRCCS, Milano), M Amigoni (Regione Lombardia, Direzione Generale
Sanit), G Micieli (IRCCS Fondazione Istituto Neurologico C Mondino,
Pavia), R Sterzi (SC Neurologia, Ospedale Niguarda CaGranda,
Milano), V Toso (Divisione Neurologica, ULSS 06 Veneto, Vicenza),
C Argentino (Dipartimento Scienze Neurologiche, Universit La
Sapienza, Roma).
Regional SupervisorsLombardia: G Micieli (IRCCS Fondazione Istituti
Neurologico C Mondino di Pavia), I Santilli (Ospedale Niguarda
CaGranda di Milano), M Stramba-Badiale (Istituto Auxologico Italiano
Ospedale San Luca IRCCS di Milano), G P Anzola (Clinica Neurologica
Ospedale SantOrsola Fatebenefratelli di Brescia). Veneto: V Toso (ULSS
06 Veneto Ospedale Civile San Bortolo di Vicenza Regione Veneto).
Friuli Venezia Giulia: F Chiodo (Grandi Azienda Ospedaliera Ospedali
Riuniti di Trieste). Liguria: M Del Sette (Dipartimento di Scienze
NeurologicheUniversit di Genova Emilia). Romagna: D Guidetti
(Azienda Ospedaliera di Reggio Emilia). Toscana: P Nencini (Azienda
Ospedaliera Ospedale Careggi di Firenze). Lazio: C Argentino
(Policlinico Umberto I Universit La Sapienza di Roma).
Researchers who did the ward auditM. Alecci, S Cesi, F Chiodo Grandi,
A Cramaro, E Montinaro, E Morandi, M Orsenigo, A Pisanello,
M Sommavilla, A Sugo, L Tancredi, E Zambrelli, P Zavarise.
Clinicians who gathered data from clinical recordsA Bersano, C Agosti,
A M Basile, E Biagioni, S Cesi, P Ferrari, S Mantovani, M Orsenigo,
S Nappini, A Poggesi, N Reale, A Rocchetto, C Sarti, L Tancredi,
A Tegani, A Tinti, P Tosi, V Vergani.
Assessors of patient follow-upH Aabid, A Argentino, E Ballabio,
G Bellatorre, S Borsa, P Bulgarelli, A Cattaneo, I Fabietti, L Facoetti,
D Gambaretti, M Groppo, E Impellizzeri, I Limosani, M Limardo,
S Lucarelli, D Messina, A Moriggia, E Moro, F Ortolano, G Rossetti,
P Stradiotti, V Tosca, P Tosi, R Virgilio.
Advisory boardW Hache (Department of Neurology University
Hospital of Heidelberg, Germany), K Asplund (Department of Medicine
University Hospital of Umea, Sweden), P Langhorne (Academic Section
of Geriatric Medicine Royal Inrmary of Glasgow, UK).
Statistical adviserA Morabito (Cattedra di Statistica Medica, Facolt di
Medicina e Chirurgia, Universit degli Studi di Milano, Milano).
PROSIT collaborating hospitals
LombardiaVarese Ospedale di Circolo (VA), Busto Arsizio (VA),
Gallarate (VA), Saronno (VA), Somma Lombardo (VA), Tradate (VA),
Angera (VA), Como Ospedale S Anna (CO), Como Ospedale G.
Valduce (CO), Cant (CO), Mariano Comense (CO), Erba (CO),
Gravedona (CO), Lecco Ospedale A Manzoni (LC), Lecco Ospedale
di Circolo (LC), Merate (LC), Sondrio (SO), Sondalo (SO), Milano
Ospedale Fatebenefratelli (MI), Milano Ospedale Niguarda (MI),
Milano Ospedale S Carlo (MI),Milano Ospedale S Paolo (MI), Milano
Ospedale L Sacco (MI), Milano Ospedale Maggiore (MI), Milano
Istituto C Besta (MI), Milano S Raaele (MI), Milano Auxologico (MI),
Monza Ospedale S Gerardo (MI), Sesto San Giovanni (MI), Cinisello
Balsamo (MI), Garbagnate Milanese (MI), Desio (MI), Carate Brianza
(MI), Melzo (MI), Vizzolo Predabissi (MI), Rho (MI), Magenta (MI),
Abbiategrasso (MI), Vaprio dAdda (MI), Vimercate (MI), Giussano
304
(MI), Cernusco Sul Naviglio(MI), Milano Santa Rita (MI), Milano Villa
Turro (MI), San Donato Milanese (MI), Legnano (MI), Lodi (LO),
Codogno (LO), Bergamo Ospedali Riuniti (BG), Treviglio (BG), San
Giovanni Bianco (BG), Alzano Lombardo (BG), Clusone (BG), Lovere
(BG), Romano di Lombardia (BG), Seriate (BG), Osio Sotto (BG),
Ponte San Pietro (BG), San Pellegrino Terme (BG), Brescia Spedali
Civili (BS), Brescia Ospedale S Orsola (BS), Desenzano del Garda (BS),
Gavardo (BS), Leno (BS), Montichiari (BS), Gardone Val Trompia (BS),
Iseo (BS), Chiari (BS), Manerbio (BS), Esine ( BS), Gussago (BS), Pavia
Istituto C. Mondino (PV), Pavia Ospedale S. Matteo (PV), Broni (PV),
Mede (PV), Vigevano (PV), Voghera (PV), Mortara ( PV), Cremona
(CR), Crema (CR), Casalmaggiore (CR), Mantova Ospedale C Poma
(MN), Asola (MN), Suzzara (MN), Pieve di Coriano (MN), Castiglione
delle Stiviere (MN). LiguriaSavona(SV), Albenga (SV), Pietra Ligure
(SV),Genova Ospedale Galliera (GE), Genova Ospedale P Micone (GE),
Genova Villa Scassi (GE), Genova Ospedale S Martino (GE), Lavagna
(GE), La Spezia (SP), Sarzana (SP), Imperia(IM),Sanremo (IM),
Bordighera (IM). VenetoVerona Ospedale Civile (VR), Verona
Ospedale Policlinico (VR), Legnago (VR), Villafranca (VR), Peschiera
del Garda (VR), Negrar (VR), Vicenza (VI), Arzignano (VI), Bassano
del Grappa (VI), Noventa Vicentina (VI), Schio (VI), Thiene (VI),
Valdagno (VI), Belluno (BL), Feltre (BL), Pieve di Cadore (BL),
Treviso(TV), Castelfranco Veneto (TV), Montebelluna (TV), Oderzo
(TV), Venezia (VE), Dolo (VE), Mirano (VE), Mestre-Venezia (VE),
Portogruaro (VE), San Don di Piave (VE,) Jesolo (VE), Padova Azienda
Ospedaliera (PD), Padova Presidio Geriatrico (PD), Camposampiero
(PD), Monselice (PD), Cittadella (PD, Abano Terme (PD), Rovigo (RO),
Adria (RO), Porto Viro (RO), Trecenta (RO). Friuli Venezia Giulia
Trieste Ospedale di Cattinara(TS), Trieste Ospedale Santorio(TS),
Udine (UD), Gemona del Friuli (UD), Latisana (UD), Palmanova (UD),
S Daniele del Friuli (UD), Tolmezzo (UD), Gorizia (GO), Monfalcone
(GO), Pordenone (PN), S Vito al Tagliamento (PN), Spilimbergo (PN).
Emilia RomagnaPiacenza (PC), Castel San Giovanni (PC),
Fiorenzuola DArda (PC), Parma Stuard (PR), Parma (PR), Fidenza
(PR), Reggio Emilia (RE), Guastalla (RE), Castelnuovo Ne Monti (RE),
Modena Policlinico (MO), Modena Ospedale Civile (MO), Carpi (MO),
Castelfranco Emilia (MO), Sassuolo (MO), Bologna Policlinico (BO),
Bologna Ospedale S Orsola (BO), Bologna Ospedale Maggiore (BO),
Bazzano (BO), Porretta Terme (BO), Imola (BO), Budrio (BO),
Bentivoglio (BO), S Giovanni in Persicelo (BO), Ferrara (FE), Cento
(FE), Argenta (FE), Comacchio(FE), Copparo (FE), Ravenna (RA),
Lugo di Ravenna (RA), Faenza (RA), Forli (FO), Forlimpopoli (FO),
Cesena Ospedale Bufalini (FO), Cesena C C Malatesta (FO), Cesena
C C S Lorenzino (FO), Rimini (RN), Cattolica (RN), Riccione (RN).
ToscanaCarrara (MS), Massa (MS), Fivizzano (MS), Pontremoli (MS),
Lucca (LU), Castelnuovo di Garfagnana (LU), Barga(LU), Pietrasanta
(LU), Viareggio (LU), Seravezza (LU), Pistoia (PT), Pescia (PT),
Pontedera (PI), S Miniato (PI), Cecina (LI), Livorno (LI), Piombino
(LI), Portoferraio (LI), Siena (SI), Poggibonsi (SI), Montepulciano (SI),
Bibbiena (AR), San Sepolcro (AR), Cortona (AR), Arezzo (AR), S
Giovanni Valdarno (AR), Montevarchi (AR), Massa Marittima (GR),
Obetello (GR), Grosseto (GR), Pisa (PI), Firenze Ospedale Nuovo (FI),
Firenze Ospedale S Maria (FI), Firenze Ospedale Careggi (FI), Borgo
S Lorenzo (FI, Prato (FI), Bagno a Ripoli (FI), Figline Valdarno (FI),
Castelorentino (FI), Empoli (FI). LazioViterbo (VT),
Acquapendente (VT), Civita Castellana (VT), Montescone (VT),
Ronciglione (VT), Tarquinia (VT), Magliano Sabina (VT), Rieti(RI),
Roma Ospedale S Giacomo (RM), Roma Ospedale CTO (RM), Roma
Ospedale Giovan Battista Grassi (RM), Roma Ospedale S Eugenio
(RM), Roma Ospedale Fatebenefratelli S Pietro (RM), Roma Ospedale
Fatebenefratelli S Giovanni (RM), Roma Ospedale S Carlo (RM),
Roma Ospedale Israelitico (RM), Roma C C Villa Aurora (RM), Roma
Life Hospital (RM), Roma Ospedale S Camillo (RM), Roma Ospedale
S Giovanni (RM), Roma Ospedale S Filippo (RM), Roma Ospedale
S Pertini (RM), Roma Ospedale Policlinico (RM), Genzano di Roma
(RM), Anzio (RM), Albano Laziale (RM), Civitavecchia (RM),
Colleferro (RM), Frascati (RM), Marino (RM), Monterotondo (RM),
Subiaco (RM), Tivoli (RM), Velletri (RM), Bracciano (RM), Palombara
Sabina (RM), Valmontone (RM), Pomezia (RM), Latina (LT), Formia
(LT), Aprilia (LT), Frosinone (FR), Alatri (FR), Anagni (FR), Ceccano
(FR), Pontecorvo (FR), Sora (FR), Cassino (FR).
www.thelancet.com Vol 369 January 27, 2007

Contributors
L Candelise contributed to the design of the study, coordination, data
collection and management, statistical analysis, interpretation of the
results, and writing of the manuscript. M Gattinoni contributed to
coordination, data collection and management, statistical analysis, and
writing of the manuscript. A Bersano participated in the coordination,
data collection and management, interpretation of the results, and
writing of the manuscript. G Micieli took part in the design of the study,
coordination, interpretation of the results, and writing of the
manuscript. R Sterzi contributed to the design of the study,
coordination, interpretation of the results, and writing of the
manuscript. A. Morabito contributed to the design of the study, and
statistical aspects. All authors saw and approved the nal version of the
manuscript.
Conict of interest statement
We declare that we have no conict of interest.
Acknowledgments
We thank the directors and the administrative sta of the hospitals and
the participating wards directors for their useful collaboration.
References
1 Stroke Unit Trialists Collaboration. Organised inpatient (stroke
unit) care for stroke. Cochrane Database Syst Rev 2001; 3: CD000197.
2 Brainin M, Bornstein N, Boysen G, Demarin V. Acute neurological
stroke care in Europe: results of the European Stroke care inventory.
Eur J Neurol 2000; 7: 510.
3 Roine RO, Kaste M. Published guidelines : updates and synthesis.
In: Bogousslavsky K, ed. Acute stroke treatment, 2nd edn. London:
Martin Dumitz, 2003; 26168.
4 Alberts MJ, Latchaw RE, Selman WR, et al. Recommendations for
comprehensive stroke centers. A consensus statement from Brain
Attack Coalition. Stroke 2005; 36: 1597618.
5 Norrving B. Organised stroke care. The core of eective stroke care
provision. Stroke 2005; 36: 161618.
6 Candelise L, Micieli G, Sterzi R, Morabito A, on behalf of the
PROSIT collaborators. Stroke unit and general wards in seven
Italian regions: the PROSIT study. Neurol Sci 2005; 26: 8194.
7 Spolaore P, Brocco S, Fedeli U, et al. Measuring accuracy of
discharge diagnoses for a region-wide surveillance of hospitalized
strokes. Stroke 2005; 36: 103134.
8 Sterzi R, Micieli G, Candelise L, on behalf of the PROSIT
collaborators. Assessment of regional acute stroke unit care in Italy:
the PROSIT study. Cerebrovasc Dis 2003; 15 (suppl 1): 1618.
9 van Swieten JC, Koudstaal PJ, Visser MC, Schouten HJ, van Gijn J.
Interobserver agreement for assessment of handicap in stroke
patients. Stroke 1998; 19: 60407.
10 Candelise L, Pinardi G, Aritzu E, Musicco M. Telephone interview
for stroke outcome assessment. Cerebrovasc Dis 1994; 4: 34143.
www.thelancet.com Vol 369 January 27, 2007
Articles
11 Hayes RJ, Bennett S. Simple sample size calculation for cluster-
randomized trials. Int J Epidemiol 1999; 28: 31926.
12 Kerry SM, Bland JM. The intracluster correlation coecient in
cluster randomisation. BMJ 1998; 316: 145560.
13 Ronning OM, Guldvog B. Stroke units versus general medical
wards. II: neurological decits and activities of daily living: a quasi-
randomised controlled trial. Stroke 1998; 29: 58690.
14 Stegmayr B, Asplund K, Hulter-Asperg K, et al. Stroke units in their
natural habitat. Can results of randomized trials be reproduced in
routine clinical practice? Stroke 1999; 30: 70914.
15 Glader EL, Stegmayer B, Johansson L, Hulter-Asberg K, Wester PO.
Dierences in long-term outcome between patients treated in
stroke units and in general wards. A 2-year follow-up of stroke
patients in Sweden. Stroke 2001; 32: 211430.
16 Rudd AG, Homan A, Irwin P, Lowe D, Pearson MG. Stroke unit
care and outcome. Results from the 2001 National Sentinel Audit
of Stoke (England, Wales, and Northern Ireland). Stroke 2005; 36:
10306.
17 Douglas VC, Tong DC, Gillum LA, et al. Do the Brain Attack
Coalitions criteria for stroke centers improve care for ischemic
stroke? Neurology 2005; 64: 42227.
18 Indredavik B, Bakke F, Slordahl SA, Rokseth R, Haheim LI. Stroke
unit treatment: 10 year follow-up. Stroke 1999; 30: 152427.
19 Kaste M, Palomaki H, Sarna S. Where and how should elderly
stroke patients be treated. A randomised trial. Stroke 1995; 26:
24953.
20 Cadilhac DA, Ibrahim J, Pearce DC, et al. Multicenter comparison
of processes of care between stroke units and conventional care
wards in Australia. Stroke 2004; 35: 103540.
21 Evans A, Perez I, Harraf F, et al. Can dierences in management
processes explain dierent outcomes between stroke unit and
stroke-team care? Lancet 2001; 358: 158292.
22 Dennis M. Stroke services: the good, the bad and the
J R Coll Physicians Lond 2000; 34: 9296.
23 Kapral MK, Laupacis A, Phillips SJ, et al. Stroke care delivery in
institutions participating in the Registry of the Canadian Stroke
Network. Stroke 2004; 35: 175662.
24 Sulter G, Elting JW, Langedijk M, Maurits NM, De Keyser J.
Admitting acute ischemic stroke patients to a stroke care
monitoring unit versus a conventional stroke unit. A randomised
pilot study. Stroke 2003; 34: 10104.
25 Cavallini A, Micieli G, Marcheselli S, Quaglini S. Role of
monitoring in management of acute ischemic stroke patients.
Stroke 2003; 34: 15.
26 Indredavik B, Bakke F, Slordhal SA, Rokseth R, Haheim LL.
Treatment in combined acute and rehabilitation stroke unit: which
aspects are most important? Stroke 1999; 30: 91723.
305